Exploring risk factors for cervical lymph node metastasis in papillary thyroid microcarcinoma: construction of a novel population-based predictive model

Table 3 Predictive performance comparison of the 5 types of machine learning classifiers in the training set and the validation set, (Mean ± SD)

ML classifiers	Accuracy(95%CI)	Accuracy(95%CI)	Sensitivity(95%CI)	Specificity(95%CI)	NPV (95%CI)
Training sets
XGBoost	0.782(0.772–0.792)^a	0.682(0.666–0.698)	0.746(0.726–0.767)	0.671(0.649–0.692)	0.957(0.955–0.959)
RF	0.679(0.635–0.724)	0.679(0.635–0.724)	0.748(0.713–0.783)	0.640(0.595–0.684)	0.953(0.945–0.961)
AdaBoost	0.648(0.612–0.685)	0.648(0.612–0.685)	0.785(0.735–0.835)	0.590(0.544–0.636)	0.953(0.945–0.962)
GNB	0.662(0.655–0.669)	0.662(0.655–0.669)	0.692(0.658–0.726)	0.683(0.645–0.722)	0.953(0.949–0.956)
MLP	0.683(0.670–0.695)	0.683(0.670–0.695)	0.711(0.658–0.764)	0.683(0.621–0.746)	0.953(0.948–0.958)
Validation sets
XGBoost	0.777(0.757–0.797)^a	0.678(0.663–0.694)	0.748(0.717–0.780)	0.660(0.614–0.706)	0.956(0949–0.962)
RF	0.679(0.635–0.724)	0.679(0.635–0.724)	0.748(0.713–0.783)	0.640(0.595–0.684)	0.953(0.945–0.961)
AdaBoost	0.648(0.612–0.685)	0.648(0.612–0.685)	0.785(0.735–0.835)	0.590(0.544–0.636)	0.953(0.945–0.962)
GNB	0.662(0.655–0.669)	0.662(0.655–0.669)	0.692(0.658–0.726)	0.683(0.645–0.722)	0.953(0.949–0.956)
MLP	0.683(0.670–0.695)	0.683(0.670–0.695)	0.711(0.658–0.764)	0.683(0.621–0.746)	0.953(0.948–0.958)

Note: ^aindicated that the best performance of the ML classifier in the training set and validation sets was XGBoost (Ranked according to AUC)
Abbreviation:ML Machine learning, XGBoost Extreme gradient boosting, RF Random forest, AdaBoost Adaptive boosting, GNB Gaussian naive Bayes, MLP Multilayer perceptron, AUROC Area under the receiver operating characteristic curve, NPV Negative predictive value, CI Confidence interval

ISSN: 1472-6823